Observed Signal · May 5, 2026 · Technical Release · Source: OpenAI Blog · Impact: 4/5 · Sentiment: Positive
OpenAI and partners publish MRC supercomputer networking protocol
OpenAI, together with AMD, Broadcom, Intel, Microsoft and NVIDIA, has published MRC (Multipath Reliable Connection), a new networking protocol for large-scale AI training. Released through the Open Compute Project (OCP), MRC extends RoCE and uses SRv6-based source routing, multi-plane topologies, adaptive packet spraying and packet trimming to improve throughput, resilience and predictability in large GPU clusters. OpenAI reports MRC is deployed across its largest NVIDIA GB200 supercomputers — including a Stargate site on Oracle Cloud Infrastructure (OCI) in Abilene, Texas and Microsoft’s Fairwater deployments — and claims the design allows two-tier switch fabrics to connect over 100,000 GPUs with reduced power, cost and failure impact. The specification and a co-authored paper were published to enable industry adoption and broader cross-vendor deployments.
Open-sourced network protocol and specification from a major AI platform (OpenAI) with cross-industry partners can materially improve large-scale model training efficiency, resilience and hardware utilization—affecting AI infrastructure and deployment strategies.
Track NVIDIA Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- OpenAI partnered with AMD, Broadcom, Intel, Microsoft and NVIDIA to develop MRC (Multipath Reliable Connection).
- OpenAI released the MRC specification through the Open Compute Project (OCP).
- MRC extends RoCE and uses SRv6-based source routing, multi-plane networks, adaptive packet spraying and packet trimming.
- MRC is deployed across OpenAI’s largest NVIDIA GB200 supercomputers, including a Stargate site on Oracle Cloud Infrastructure (OCI) in Abilene, Texas and Microsoft’s Fairwater supercomputers.
- OpenAI claims MRC enables two-tier switch fabrics to connect over 100,000 GPUs while reducing power, components, and routing-induced training interruptions.
Connected Companies & Entities
6 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
AMD partners with science ministry for open-weight AI
AMD confirmed an alliance with a ministry of science to build an open-weight AI ecosystem, according to an MSN report published on 2026-08-11. The company already publishes fully open models (weights, training data and recipes) — notably OLMo (2024) and Instella (2025) — trained on AMD Instinct MI300X accelerators. AMD positions its open-source ROCm GPU software stack as an alternative to Nvidia's largely closed CUDA ecosystem and in 2025 expanded its physical AI infrastructure by acquiring ZT Systems. The MSN coverage did not disclose the country, investment amounts or timeline for the ministry partnership. AMD's open models are available from huggingface.co/amd. The alliance aims to give governments, universities and researchers auditable, locally runnable models and to promote ROCm adoption in public procurement.
OpenAI and Broadcom Team Up for AI Accelerator Revolution
OpenAI and Broadcom announced a multi‑year strategic collaboration to co-develop and deploy 10 gigawatts of custom AI accelerators and associated network systems. OpenAI will design the accelerators and systems, and Broadcom will provide Ethernet, PCIe and optical connectivity solutions and deploy racks of the accelerators. The deployment is targeted to begin in the second half of 2026 and to complete by the end of 2029. The companies have signed a term sheet for racks incorporating OpenAI’s accelerators and Broadcom networking solutions, with deployments across OpenAI facilities and partner data centers. The announcement emphasizes custom accelerators and Ethernet-based scale-up and scale-out networking as key elements for next-generation, power-efficient AI clusters. OpenAI stated it has over 800 million weekly active users.
OpenAI Teams Up with Cerebras for AI Innovation
OpenAI announced a partnership with Cerebras to add 750 MW of ultra low-latency AI compute to OpenAI’s platform. Cerebras’ systems use a single large chip combining compute, memory and bandwidth to accelerate inference. OpenAI said the capacity will be integrated into its inference stack in phases across workloads and will come online in multiple tranches through 2028. OpenAI’s Sachin Katti and Cerebras co-founder and CEO Andrew Feldman are quoted describing the deal’s aim to enable faster, more natural real-time AI interactions and scale real-time inference to more users. The announcement was published by OpenAI on January 14, 2026.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.